1

Scalable Video Multicast in Hybrid 3G/Ad-hocNetworks

Sha Hua, Yang Guo, Member, IEEE, Yong Liu, Member, IEEE, Hang Liu, Member, IEEE, and Shivendra S.Panwar, Fellow, IEEE

Abstract—Mobile video broadcasting service, or mobile TV, isexpected to become a popular application for 3G wireless networkoperators. Most existing solutions for video Broadcast MulticastServices (BCMCS) in 3G networks employ a single transmissionrate to cover all viewers. The system-wide video quality of thecell is therefore throttled by a few viewers close to the boundary,and is far from reaching the social-optimum allowed by the radioresources available at the base station. In this paper, we proposea novel scalable video broadcast/multicast solution, SV-BCMCS,that efficiently integrates scalable video coding, 3G broadcastand ad-hoc forwarding to balance the system-wide and worst-case video quality of all viewers at 3G cell. We solve the optimalresource allocation problem in SV-BCMCS and develop practicalhelper discovery and relay routing algorithms. Moreover, weanalytically study the gain of using ad-hoc relay, in terms ofusers’ effective distance to the base station. Through extensivereal video sequence driven simulations, we show that SV-BCMCSsignificantly improves the system-wide perceived video quality.The users’ average PSNR increases by as much as 1.70 dB withslight quality degradation for the few users close to the 3G cellboundary.

Index Terms—Scalable Video Coding, Resource Allocation, Ad-hoc Video Relay,

EDICS Category

5-WRLS (Wireless Multimedia Communication)

I. INTRODUCTION

User demands for content-rich multimedia are driving muchof the innovation in wireline and wireless networks. Watchinga movie or a live TV show on their cell phones, at anytimeand at any place, is an attractive application to many users.Mobile video broadcasting service, or mobile TV, is expectedto become a popular application for 3G network operators.The service is currently operational, mainly in the unicastmode, with individual viewers assigned to dedicated radiochannels. However, a unicast-based solution is not scalable. A

This work is supported by the New York State Center for AdvancedTechnology in Telecommunications (CATT) and the Wireless Internet Centerfor Advanced Technology (WICAT).

S. Hua, Y. Liu and S. S. Panwar are with the Department of Electrical andComputer Engineering, Polytechnic Institute of New York University, Brook-lyn, NY 11201 USA (e-mails: shua01@students.poly.edu; yongliu@poly.edu;panwar@catt.poly.edu).

Y. Guo is with the Service Infrastructure Research Department at BellLabs, Alcatel-Lucent, Crawford Hill, Holmdel, NJ 07733 USA (e-mail:Yang.Guo@alcatel-lucent.com)

H. Liu is with the Technicolor Corporate Research, Princeton, NJ 08540USA (e-mail: liu1999@ieee.org)

Copyright (c) 2010 IEEE. Personal use of this material is permitted.However, permission to use this material for any other purposes must beobtained from the IEEE by sending a request to pubs-permissions@ieee.org.

Broadcast/Multicast service over cellular networks is a moreefficient solution with the benefits of low infrastructure cost,simplicity in integration with existing voice/data services, andstrong interactivity support. Thus, it is a significant part of3G cellular service. For instance, Broadcast/Multicast Ser-vices (BCMCS) [1], [2] has become a part of the standardsin the Third Generation Partnership Project 2 (3GPP2) [3]for providing broadcast/multicast service in the CDMA2000setting. However, most existing BCMCS solutions employ asingle transmission rate to cover all viewers, regardless of theirlocations in the cell. Such a design is sub-optimal. Viewersclose to the base-station are significantly “slowed down” byviewers close to the cell boundary. The system-wide perceivedvideo quality is far from reaching the social-optimum.

In this paper, we propose a novel scalable video broad-cast/multicast solution, SV-BCMCS, that efficiently integratesscalable video coding, 3G broadcast and ad-hoc forwardingto achieve the optimal trade-off between the system-wide andworst-case video quality perceived by all viewers in the cell.In our solution, video is encoded into one base layer andmultiple enhancement layers using Scalable Video Coding(SVC) [4]. Different layers are broadcast at different ratesto cover viewers at different ranges. To provide the basicvideo service to all viewers, the base layer is always broadcastto the entire cell. The channel resources/bandwidth of theenhancement layers are optimally allocated to maximize thesystem-wide video quality given the locations of the viewersand radio resources available at the base station. In addition,we allow viewers to forward enhancement layers to each otherusing short-hop and high-rate ad-hoc connections. In summary,the major contribution of this paper is as follows:

1) We study the optimal resource allocation problem forscalable video multicast in 3G networks. We showthat the system-wide video quality can be significantlyincreased by jointly assigning the channel resources forenhancement layers. Our solution strikes a good balancebetween the average and worst-case performance for allviewers in the cell.

2) For ad-hoc video forwarding, we design an efficienthelper discovery scheme for viewers to obtain additionalenhancement layers from their ad-hoc neighbors a fewhops away. Also a multi-hop relay routing schemeis designed to exploit the broadcast nature of ad-hoctransmissions and eliminate redundant video relays fromhelpers to their receivers.

3) We develop analytical models to study the expected

2

gain of few-hop ad-hoc video relays under random userdistribution in a 3G cell. An ad-hoc relay shortens theeffective distance of the user to the base station, whichis crucial to the throughput improvement of wirelesstransmission. Based on the models, we give the math-ematical solution for the effective distance gain. Theanalysis underpins our protocol design from a theoreticalviewpoint.

4) We selected three representative video sequences, andconducted video trace driven simulations using OPNET.We systematically evaluated the performance improve-ment attributed to several important factors, includingnode density, the number of relay hops, user mobilityand coding rates of video layers. Simulations show thatSV-BCMCS significantly improves the video qualityperceived by users in practical 3G/ad-hoc hybrid net-works.

The rest of the paper is organized as follows. Related workis described in Section II. The SV-BCMCS architecture isfirst introduced in Section III. We then formulate the optimallayered video broadcast problem and present a dynamic pro-gramming algorithm to solve it. Following that, we develop thehelper discovery and multiple-hop relay routing algorithms.We develop analytical models to study the expected gain offew-hop ad-hoc video relay in Section IV. Video trace drivensimulation results are shown and discussed in Section V. Thepaper is concluded in Section VI.

II. RELATED WORK

Using ad-hoc links to help data transmissions in 3G net-works has been studied by several research groups in thepast. In [5], a Unified Cellular and Ad-hoc Network (UCAN)architecture for enhancing cell throughput has been proposed,which is the starting point of our work. Clients with poorchannel quality select clients with better channel quality astheir proxies to receive data from the 3G base station. Apacket to a client is first sent by the base station to its proxynode through a 3G channel. The proxy node then forwardsthe packet to the client through an ad-hoc network composedof other mobile clients and IEEE 802.11 wireless links. In[6], the authors discovered the reason for the inefficiencythat arises when using an ad-hoc peer-to-peer network as-isin a cellular system. Then they proposed two approaches toimprove the performance, one is to leverage assistance fromthe base station, and the other is to leverage the relayingcapability of multi-homed hosts.

While the above articles are focused on the unicast datatransmission in cellular networks, ad-hoc transmission can bealso employed to improve the performance of 3G Broadcast/-Multicast Services. Based on the UCAN, Park and Kasera [7]developed a new proxy discovery algorithm for the cellularmulticast receivers. The effect of ad-hoc path interferenceis taken into consideration. ICAM [8] developed a close-to-optimal algorithm for the construction of the multicast forestfor the integrated cellular and ad-hoc network. Sinkar et al. [9]proposed a novel method to provide QoS support by usingan ad-hoc assistant network to recover the loss of multicast

data in the cellular network. However, the aforementionedworks did not make use of SVC coded streams, which allowcellular operators to flexibly select the operating point so asto strike the right balance between the system-wide aggregateperformance and individual users’ perceived video quality.

Layered video multicasting combined with adaptive modu-lation and coding to maximize the video quality in an infras-tructure based wireless network is studied in [10]–[12]. Theydiffer in the formulation of the optimal resource allocationproblem. Peilong et al. [13] and Donglin et al. [14] extendedthis idea into multi-carrier and cognitive radio scenarios. Noneof them, however, takes the advantage of the assistance ofad-hoc networks. Our work is also relevant to the studies ofscalable video transmission over pure ad-hoc networks suchas [15]. The authors in [15] consider the scenario of scalablevideo streaming from multiple sources to multiple users. A dis-tributed scheme that optimizes the rate-distortion is introduced.In contrast, this paper studies the scalable video streaming inthe 3G networks with the assistance of ad-hoc network formedamong users’ mobile devices. The optimization problem issolved at the 3G base station. The distributed relay routingprotocol is designed to locate the helping devices. Finally, thecurrent paper substantially extends the preliminary version ofour results which appeared in [16]. The analytical models aredeveloped to study the expected gain of ad-hoc video relaysunder a random user distribution. PSNR, instead of receivedvideo rate, is used as the target function in the optimizationproblem and in the simulation experiments, which betterreflects the users’ perceived video quality. In addition, realvideo trace driven simulations are conducted to evaluate thesystem performance.

III. SCALABLE VIDEO BROADCAST/MULTICAST SERVICE(SV-BCMCS) OVER A HYBRID NETWORK

In this section, we start with the introduction of the SV-BCMCS architecture. We then formulate and solve the optimalresource allocation problem for the base station to broadcastscalable video with the aid of an ad-hoc network. Finally,we develop the helper discovery and layered video relayrouting algorithms to explore the performance improvementintroduced by ad-hoc connections between viewers.

A. System Architecture

In SV-BCMCS, through SVC coding, video is encoded intoone base layer and multiple enhancement layers. Viewers whoreceive the base layer can view the video with the minimumquality. The video quality improves as the number of receivedlayers increases. An enhancement layer can be decoded if andonly if all enhancement layers below it are received. Themulticast radio channel of the base station is divided intomultiple sub-channels. Here a sub-channel stands for eithera temporal or frequency division of the bandwidth. Differentlayers of video are broadcast using different sub-channels withdifferent coverage ranges. To maintain the minimum qualityfor all viewers, the base layer is always broadcast using asub-channel to cover the entire cell. To address the decodingdependency of upper layers on lower layers, the broadcast

3

range of lower layers cannot be shorter than that of the higherlayers.

Figure 1 depicts the system architecture of SV-BCMCS aswith eleven multicast users and three layers of SVC video,denoted by L1, L2, and L3. The base station broadcasts threelayers using three sub-channels with their respective coverageareas shown in the figure. All the users receive L1 from thebase station directly, while four users receive L2 and twousers receive L2 and L3 as well. With an ad-hoc network,the coverage of enhancement layers is extended further. As anexample, user a is in the coverage of L3 and user b is in thecoverage of L2. User a relays L3 to user b, who then relaysL3 to users c and d. Meanwhile, user b relays L2 to users cand d. Effectively, all four users a, b, c, and d receive all threelayers through the combinations of base station broadcast andad-hoc relays.

Fig. 1. Architecture of SV-BCMCS over a hybrid network (assuming threelayers of video content).

The key design questions of the SV-BCMCS architectureare:

1) How to allocate the radio resources among sub-channelsto different video layers to strike the right balance be-tween system-wide and worst-case video quality amongall users?

2) How to design an efficient helper discovery and relayrouting protocol to maximize the gain of ad-hoc videoforwarding?

We examine these questions through analysis and simulationsin the following sections. The key notation used in this paperare shown in Table I.

B. Optimal Resouce Allocation in Layered Video Multicast

Our objective for radio resource allocation is to maximizethe aggregate user perceived video quality while providing thebase-line minimum quality service for all users. The perceivedvideo quality can be measured by PSNR (Peak Signal-to-NoiseRatio) or distortion, with PSNR = 10 log10(M2

I /D) (D isthe distortion represented by the MSE (Mean Square Error)between the original image and the reconstructed image, andMI is the maximum pixel value, typically set to be 255).The PSNR or distortion of a video sequence is the averageof the corresponding measurements over all images in thesame video sequence. Modeling the distortion or the PSNRas the function of the user’s received data rate is an ongoing

TABLE INOTATION USED IN THIS PAPER

ri selected transmission rate (PHY mode) for layer ivideo content

pi allocated fraction of channel for tansmission of layeri video content in 3G system

ni number of multicast users that can receive layer ivideo/content in 3G domain

Ri encoding rate of an individual layer i.L total number of layersN total number of multicast users in the 3G domainN the set of multicast users in the 3G domain. |N | = NΦ set of all possible transmission rates (or PHY modes)Utotal aggregate utility of users in the 3G domaind distance of the user to the base stationrt ad-hoc transmission radius/range for each userG random variable of effective distance gainSc total area of the entire 3G cellD the radius of the 3G cell

research topic. For example, in [17], distortion is modeled as acontinuous function of video rate. In [15], the distortion withSVC is modeled as discrete values depending on the numberof layers received by the user. In [18] and [19], PSNR ismodeled as a linear/piece-wise linear function of the video ratewith SVC. Here we employ a general non-decreasing utilityfunction U(Rrec) in the optimization formulation, with Rrec

being a user’s receiving data rate. In Section V, we replacethe general utility function by a video sequence’s actual PSNRvalues.

Assume there are L video layers, and the video rate of eachlayer is a constant Ri, 1 ≤ i ≤ L. The broadcast channel is di-vided into L sub-channels through time-division multiplexing.Each sub-channel can operate at one of the available BCMCSPHY modes. Each PHY mode has a constant data transmissionrate and a corresponding coverage range. Layer i is transmittedusing sub-channel i. Let pi be the time fraction allocated tosub-channel i, and ri be the actual transmission rate, or thePHY mode, employed by sub-channel i. In practice, to supportthe video rate Ri of layer i, ri ·pi ≥ Ri. We let pi = Ri

riin our

formulation. In addition, the summation of the time fractionsmust be less than one

∑Li=1 pi ≤ 1.

Suppose ni multicast users can receive the i-th layer video.Therefore ni−ni−1 users, 1 ≤ i < L, receive i layers of video,while nL users receive all L layers. The aggregate utility forall users is:

Utotal =∑i∈N

U(Rreci )

= (n1 − n2) · U(R1) + · · ·+ (nj − nj+1) · U(

j∑i=1

Ri)

+ · · ·+ nL · U(

L∑i=1

Ri)

=

L∑i=1

(ni − ni+1)U(

i∑j=1

Rj). (1)

4

with nL+1 = 0. For a fixed total number of multicast users Nin the 3G domain, maximizing the average utility of multicastusers is the same as maximizing the aggregate utility Utotalin Eqn. (1).

Next, the computation of ni is discussed. Let’s first considerthe base station transmission part. The number of users thatcan receive layer i directly from the base station is determinedby the transmission rate ri for sub-channel i (specifically,determined by the receiving SNR for the PHY mode with rateri). We can define this number of users as fBS(ri) for layer i.Due to path loss, fading, and user mobility, fBS(ri) varies withri and generally is a monotonically decreasing function of ri.Namely the higher the transmission rate of the base station,the fewer the multicast users that can achieve the receivingSNR requirement and thus correctly receive the data.

In the second step, we take the ad-hoc relay into con-sideration. In SV-BCMCS, users who cannot receive layer ifrom the base station directly may receive it through the ad-hoc network. We use fAD−HOC(ri) to denote the number ofusers obtaining the layer i video through ad-hoc relay fromother users. Thus ni = f(ri) = fBS(ri) + fAD−HOC(ri). Inthe system design, the base station gathers information aboutwhich user is getting data from which helper to compute ni.This is done in the helper discovery protocol illustrated inSection III-C.

The optimal radio resource allocation problem can beformulated as the following utility maximization problem:

max{ri}

Utotal =

L∑i=1

[f(ri)− f(ri+1)]U(

i∑j=1

Rj), (2)

subject to:

ri ≤ rj i ≤ j and i, j = 1, 2, · · ·L, (3)L∑i=1

(Riri

) ≤ 1, (4)

f(r1) = N, (5)ri ∈ Φ. (6)

The objective is to find a set of transmission rates ri forindividual layers so as to maximize the aggregate utility. Theconstraint given by (3) ensures that the coverage of lowerlayers is larger than that of the higher layers. Constraint (4)guarantees the sum of sub-channels is no greater than theoriginal channel. Constraint (5) ensures that the base layercovers the whole cell to provide basic video service to all theusers. Finally, Φ is the set of possible transmission rates (orPHY modes). Note that the traditional broadcast/multicast withone single stream is a special case of the above optimizationproblem with L = 1.

We have also designed a dynamic programming algorithm tosolve the optimization problem formulated above, which worksfor any general non-decreasing function U(·). The complexityof the algorithm is O(LK2), where K is an integer forrescaling the sub-channel allocation fractions pi. K can be10n if all pi can be expressed using at most n significantdigits. Given the page limitation, the details of the algorithm

are compiled in [20].

C. Ad-hoc Video Relay: Helper Discovery and Relay Routing

In a pure SV-BCMCS solution, users closer to the basestation will receive more enhancement layers from the basestation. They can forward those layers to users further awayfrom the base station through ad-hoc links. Ad-hoc videorelays are done in two steps: 1) each user finds a helper inits ad-hoc neighborhood to download additional enhancementlayers; 2) helpers merge download requests from their clientsand forward enhancement layers through local broadcast.

1) Greedy Helper Discovery Protocol: We design a greedyprotocol for users to find helpers. A greedy helper discoveryprotocol in the 3G and ad-hoc hybrid network was firstpresented in [5]. In that paper every node of the multicastgroup maintains a list of its neighbors, containing their IDsand the average 3G downlink data rates within a time window.Users periodically broadcast their IDs and downlink data ratesto their neighbors. Each user greedily selects a neighbor withthe highest downlink rate as its helper. Whenever a node wantsto download data from the base station, it initiates helperdiscovery by unicasting a request message to its helper. Thenthe helper will forward this message to its own helper, soon and so forth, until the ad-hoc hop limit is reached or anode with the local maximum data rate is found. The ad-hochop limit is set by a parameter called Time-To-Live (TTL).The base station will send the data to the last-hop helper. Thehelpers will forward the data in the reverse direction of helperdiscovery to the requesting node.

We employ a similar greedy helper discovery mechanism.But unlike the case considered in [5], the locations andaverage 3G downlink data rates of helpers will affect theresource allocation strategy of the base station in SV-BCMCS.In our scheme, the last-hop node in the path sends thefinal request message to the base station. Upon receivingthis message, the base station updates the 3G data rateinformation of all the nodes along the path as the last-hop node’s 3G rate, assuming the ad-hoc link throughputis much larger than the 3G data rate. After that, the basestation might resolve the coverage function as f(ri) ={Number of nodes with updated 3G rate above ri}. The op-timal broadcast strategy can then be calculated by solving theoptimization problem defined in (2).

Moreover, to facilitate efficient relay routing, a node alsoneeds to keep information about the relay requests routedthrough itself. The last-hop helper in the path also sends thefinal request message back to the initiating node. Every useralong this ad-hoc path will get a copy of this final message.

As an example, the whole process is shown in Figure 2.User C attempts to find a helper within two hops to improveits video quality. Its request goes through B to A. E and Fare ignored by C and B, since they are not the neighbor withhighest 3G rate. To this end, User A knows where user C islocated by the reverse route of the path that user C followedto find user A. User A sends the request message to the basestation to indicate that user A will act as user C’s helper usingthe relay path along user B. Meanwhile, User A also sends this

5

message (confirmation message) back to user C confirmingthat user A will act as its helper.

Fig. 2. Greedy Helper Discovery Protocol and Relay Routing Protocol. Thenumbers in parenthesis are the average 3G rates for each node. The dashedline indicates the ad-hoc neighborhood. The straight solid line is the ad-hocpath with the arrow pointing to the helper. For relay routing protocol, dashed-dotted lines represent the coverage area of each layer. 27, 18, 12 and 5 arethe physical transmission rates for layers 4, 3, 2 and 1.

Note that [5] also proposed another helper discovery pro-tocol using flooding. Instead of unicast, each node broadcaststhe request message hop by hop. This method enables a nodeto find the helper with the global maximum data rate withinthe ad-hoc hop limit range. However, considering the largeoverhead of flooding messages in the ad-hoc network, weonly adopt the greedy helper discovery protocol within theSV-BCMCS context.

2) SV-BCMCS Relay Routing Protocol: The SV-BCMCSrouting protocol runs after the greedy helper selection pro-tocol and the optimal radio resource allocation. Assumingoptimal radio resource allocation has been performed, thebase station decides to transmit the L layers with differentrates r1, r2, · · · , rL. It will broadcast this information to everynode in the cell. Moreover, in the greedy helper discoveryphase, each node obtains the information for all the relaypaths to which it belongs. The major goal of the relay routingprotocol is to maximally exploit the broadcast nature of ad-hoc transmissions and merge multiple relay requests for thesame layer on a common helper.

Essentially, each helper needs to locally determine whichreceived layers will be forwarded to its requesting neighbors.For each node n, define the set K ={all neighbors that use nas one-hop helper}, the forwarding decision will be calculateddistributedly as shown in Algorithm1.

For the receiving part, each node receives packets thatsatisfy two conditions: (i) the packets are sent from its directone-hop helper; (ii) the packets belong to a layer that the nodecannot directly receive from the base station. Otherwise thenode will discard the packets. That is, the node has no use forpackets that are within the layer to which the node belongs,or from a lower layer than the layer to which it belongs.

A relay routing example is illustrated in Figure 2. SupposeL = 4 and maximal hop number (TTL) is 2. For node B in thefigure, nodes C and D use it as a direct one-hop helper. ForD, lD = 2 (it is the layer to which node D belongs) accordingto the figure, and within 2 hops, D’s highest expected layer

Algorithm 1 Forwarding Algorithm in SV-BCMCS RoutingProtocol for Node n

1: K ={all neighbors that use n as one-hop helper }2: for k ∈ K do3: Find the highest layer lk that k can directly receive from

the base station4: Find the highest layer Lk that k can expect from any

potential helper5: end for6: lmin = min{lk, k ∈ K}7: Lmax = max{Lk, k ∈ K}8: node n broadcasts the packets between layer lmin + 1 toLmax to its one-hop neighbors.

is LD = 4. The highest expected layer is the highest layerwhich a node can expect to receive through its helpers whileconstrained by the TTL. In the same way, we can derive, lC =1 and LC = 4. Thus, for node B, lmin = min{lC , lD} = 1and Lmax = max{LC , LD} = 4. Therefore, node B willbroadcast the packets in layers 2, 3 and 4. Since node D is inlayer 2, it will receive packets from node B in layers 3 and 4only. Meanwhile, node C will receive all the packets in layers2, 3 and 4.

Note a local broadcast doesn’t use RTS/CTS exchange in apractical implementation based on IEEE 802.11. Instead, thenodes’ carrier sensing threshold can be set to a reasonablesmall value. In this way, we can reduce the number ofcollisions due to hidden terminal problem while still keep afavorable spatial reuse factor in the network.

In practice, wireless channels are error-prone and link qual-ity changes over time due to the fading and interference. Thisposes a challenge particularly to the video transmission. Dueto the use of spatial temporal prediction, a compressed video issusceptible to transmission errors. To overcome such problem,appropriate error protection mechanism is necessary in thepractical implementation. FEC (Forward Error Correction) iswidely used as an effective means to combat packet losses overwireless channel [15], [21], [22], Our relay routing protocolcan be easily extended to support FEC by letting the helpernodes decode and re-generate the parity packets for eachvideo layer before forwarding them. We will evaluate theperformance of our system with FEC in Section V.

IV. ANALYSIS OF THE GAIN OF AD-HOC RELAY

In this section, we analytically study the expected gain fromusing ad-hoc relays under a random node distribution in thecell. Through ad-hoc video relays, users receiving fewer layersof packets (users at the coverage edge/boundary) are able toobtain video/content layers that they otherwise would not orcould not receive. From the base station’s viewpoint, an ad-hocrelay shortens a user’s effective distance to the base station.As shown in Figure 2, user A relays data to user B who thenrelays the data to user C. If we assume that the bandwidth ofan ad-hoc link is much larger than the 3G multicast rate, bothuser B and C can be seen as located at the same place as userA, then have the same effective distance as user A.

6

We define the distance gain of a user as the differencebetween its original distance to the base station and its lasthelper’s distance to the base station. For example, the distancegain of user C is the difference between OC and OA,where O denotes the position of the base station. We areparticularly interested in this metric due to the fact that inwireless communications, the distance between the transmitterand the receiver fundamentally affects their transmission rate.In the following, we develop a probabilistic model to studythe distance gain due to ad-hoc relays under a random nodedistribution. Note that such distance gain is an upper boundto the case with limited ad-hoc bandwidth, which is able toshow the insight of benefits brought by ad-hoc relay.

The typical transmission rate of an ad-hoc network, suchas a network using IEEE 802.11, is much larger than the rateof a cellular network. For instance, IEEE 802.11g supportsdata rates up to 54 Mb/s. We therefore ignore the effect ofwireless interference in our analysis. The interference will beincluded in our OPNET based simulation in the next section.We also assume that the number of data relays, or relay hops,is small. Using only a small number of relays is more robustagainst user mobility, and reduces the video forwarding delay.Furthermore, a smaller number of relays also reduces thetraffic volume in the whole ad-hoc network.

Let G be the distance gain of an arbitrary user. Obviously, Gdepends on the location of the user, as well as the locations ofother users in the same cell. It is also a function of the ad-hoctransmission range, and the number of relay hops allowed.Denote by fG(·) the probability density function (pdf) ofG. We develop a model to characterize fG(·) by assumingthe users are uniformly distributed in the cell. Our approach,however, also applies to other distributions within the cell. Thelist of key notation is included in Table I. Figure 3 depicts anarbitrary user and is used to study the user’s distance gain inthe case of a one-hop and two-hop relay.

A. Distance Gain Using One-hop Relay

Assuming that the user is d distance away from the basestation, and the ad-hoc transmission radius/range is rt. Allother users falling into the transmission range of the userare potential one-hop helpers. Following the greedy helperdiscovery protocol, the user closest to the base station ischosen as the relay node. To calculate fG(g), we need tocalculate the probability that the distance gain is in a smallrange of [g, g+∆g]. As illustrated in the left part of Figure 3,the whole cell space is divided into three regions: S1, S2 andS3. Since the relay node is closer to the base station thanany other node falling into the transmission range of the user,there should be no node in S2 in the left part of Figure 3. Toachieve a distance gain of [g, g+ ∆g], there should be at leastone node falling into S3. Since the area of S3 is proportionalto ∆g, the probability that two or more nodes fall into S3 isa higher order of ∆g, and thus will be ignored. Therefore, theprobability of the distance gain in the range of [g, g + ∆g]is the probability that when we randomly drop N − 1 nodes(excluding the user under study) into the cell, one node fallsinto S3, no node falls into S2 and N − 2 nodes fall into the

remaining area S1. Based on the multinomial distribution:

fG(g) = lim∆g→0

(7)

[Pr(N − 2 nodes in S1, no node in S2, 1 node in S3)

∆g]

= lim∆g→0

(N − 1)!

(N − 2)!1!0!(q1)N−2(q2)0 q3

∆g

= lim∆g→0

(N − 1)qN−21

q3

∆g. (8)

where q1, q2, q3 are the probabilities of users located in thearea S1, S2, S3. Due to the uniform distribution of the users,qi = Si

Sc, i = 1, 2, 3, where Sc is the area of entire cell.

S2 is the overlapping area of two circles. For two circleswith a known distance d12 between their centers and with ra-dius of each circle c1 and c2, let SII(d12, c1, c2) represent theiroverlapping area. The detailed derivation of SII(d12, c1, c2) isavailable in [23]. For our case, S2 = SII(d, rt, d − g). For afixed rt, S2 is a function of g and d, so we use S2(g, d) fromthis point on. It is easy to verify that

lim∆g→0

q1 = 1− S2(g, d)

Sc= 1− SII(d, rt, d− g)

Sc(9)

and

lim∆g→0

q3

∆g· SC

=S3

∆g= −dS2(g, d)

dg=

dSII(d, rt, x)

dx|x=d−g. (10)

Consequently, we have

fG(g, d) =N − 1

Sc·(

1− SII(d, rt, d− g)

Sc

)N−2

·dSII(d, rt, x)

dx|x=d−g. (11)

The expected one-hop distance gain for a user at a distance dfrom the base station can be derived as:

µG(d) =

∫ rt

0

gfG(g, d)dg. (12)

B. Distance Gain Using Two-hop Relay

The way to derive the distance gain in the two-hop caseis similar to the one-hop case. However the two hop caseis computationally more complex. To accurately characterizethe two-hop relay gain, we need to calculate the joint densityof the distance gain of the first and second relay hop. Asillustrated in the right part of Figure 3, let g1 be the distancegain of the first-hop relay, θ be the angle between the first-hophelper and the user with the base station as the origin, g2 bethe distance gain of the second-hop relay. In a manner similarto the one-hop case, we can calculate the joint density functionf(g1, θ, g2) by calculating the multinomial distribution of N−1 nodes fall into five regions as illustrated in the right part ofFigure 3. Let ni be the number of nodes falling into areaSi. We need to calculate the multinomial probability of n1 =N − 3, n2 = n4 = 0, n3 = n5 = 1. The areas of S1∼5 arefunctions of g1, g2, d and θ.

7

Fig. 3. The one-hop(left) and two-hop(right) ad-hoc relay analysis for an arbitrary user.

The pdf of the joint distance gain can be calculated as:

f(g1, g2, θ) = lim∆g1,∆θ,∆g2→0

[(N − 1)(N − 2)

SN−1c

· SN−31 · S3

∆g2· S5

∆g1∆θ]. (13)

It can be determined that S5 = (d − g1)∆θ∆g1 and S1 =Sc−

∑5i=2 Si. Unfortunately, the calculation of S2, S3 and S4

is fairly involved. Given the page limitation, we have compiledthe details in [20].

C. Impact of ad-hoc wireless relay to user performance

With the concept of “distance gain”, we can think of moreusers moving closer to the base station and “appear to exist”within certain distance of the base station compared to thescenario with no ad-hoc relay. The benefit it brings to ourlayered video multicast is that the video layers can be receivedby more users, which have shown in the previous sections.Next we will analytically derive the number of users thateffectively move closer to the base station with the aid ofad-hoc wireless relay.

Based on the model we build up, our objective is to calculateon average how many users outside a given distance d canmove into the circle, with the aid of ad-hoc relay. If we sup-pose that different video layers are transmitted with differentranges, such increase represents the number of additional usersthat can receive a certain video layer. So it has a significantpractical meaning.

In detail, we divide the ring between a distance d andd + rt into many concentric rings, each with a width of 4.Note that rt is the range of ad-hoc transmission. One-hopad-hoc relay is considered in this case; however the approachcan be applied to the multiple hop relay scenario. N is thetotal number of multicast users in the entire cell, and D isthe radius of the cell. The average number of users in the kthring is:

Nk(4) = N · π[(d+ k4)2 − (d+ (k − 1)4)2]

πD2

= N4[(2k − 1)4+ 2d]

D2. (14)

Then, the probability that a user in the k-th ring can move

within distance d is:

pk(4) =

∫ rt

k4fG(g, d)dg

The average number of users that moves into the circle ofradius d is:

Nave =

b rt4 c∑k=1

Nk(4) · pk(4). (15)

As 4→ 0, Equation (15) can be rewritten as:

Nave(d) =

∫ rt

0

2N(d+ r)

D2

∫ rt

r

fG(g, d)dgdr. (16)

Recall in our formulation (2), with the assistance of the ad-hocnetwork, the base station can reach a larger number of usersf(ri) = fBS(ri)+fAD−HOC(ri). Now fAD−HOC(ri) can beapproximated by Nave(di), where di, the distance for certaintransmission rate ri, is discussed and derived in section III-B.

D. Numerical Results Using the Analytical ModelBased on the analytical model presented above, we numer-

ically computed the resulting distance gain and user numberincrease. Since the pdf fG(g, d) is a function of d, the userachieves different distance gain when its distance to the basestation varies. The results in this section are for different nodedensities in a 3G cell with a radius 1000 meters. We set the ad-hoc range at 100 meters. Therefore, if there are a totally 500nodes, on average each node has 500 · π1002

π10002 = 5 neighbors.Figure 4 shows how the distance gain g derived in sec-

tion IV-A and IV-B changes with d when TTL is set to oneand two. For example, when the total number of multicastusers is 700, for the user at the boundary of the cell, i.e,d = 1000, the expected one-hop and two-hop distance gainsare gTTL=1 = 62.86 and gTTL=2 = 121.61.

For the same setting with 500 multicast users, we calculatethe increase in the numbers of users at different rangesaccording to Equation (15). For d = 500, 700 and 900meters, the “number increase/original number of users” are28.57/125, 38.32/245, 48.08/405 respectively. Note withoutad-hoc, the original number of users goes proportional tod2 due to the uniform user distribution. We can observe anobvious increase as ad-hoc relay squeezes the users towardsthe base station. This explains the potential of video quality

8

100 200 300 400 500 600 700 800 900 100040

50

60

70

80

90

100

110

120

130

Distance to the Base Station (meters)

The

Ave

rage

Dis

tanc

e G

ain

(met

ers)

600 nodes, one−hop600 nodes, two−hop500 nodes, one−hop500 nodes, two−hop400 nodes, one−hop400 nodes, two−hop

Fig. 4. The one-hop and two-hop Distance Gain for nodes with differentdistance to the base station.

improvement by using ad-hoc network in the SV-BCMCSprotocol.

V. PERFORMANCE EVALUATION

In this section, the performance of SV-BCMCS is evaluatedusing OPNET based simulations. Compared to the popularnetwork simulators ns-2 and Omnet++ which are free, OPNETis a commercial network simulator using standards basedmodels [24]. The performance of SV-BCMCS is comparedwith the performance of traditional 3G BCMCS under variousscenarios. The impact of node density, node mobility, numberof relay hops and the base layer video rate is investigated. Re-sults demonstrate that SV-BCMCS consistently out-performsBCMCS with or without the aid of ad-hoc data relay.

A. Simulation Settings

1) Network Settings: SV-BCMCS is simulated using thewireless modules of OPNET modeler. It is assumed thatall multicast users/nodes have two wireless interfaces: onesupports a CDMA2000/BCMCS channel for 3G video service,and the other supports IEEE 802.11g for ad-hoc data relay.The data rate of the ad-hoc network is set to be 54 Mb/s,and the transmission power covers 100 meters. Since OPNETmodeler does not provide built-in wireless modules with dualinterfaces, the 3G downlink is simulated as if individual usersgenerate their own 3G traffic according to the experimentaldata presented in [1], [25]. The free-space path loss modelis adopted for 3G downlink channels, where the Path LossExponent (PLE) is set to be 3.52, and the received thermalnoise power is set to be -100.2dBm. Eleven PHY data ratesare supported according to the 3GPP2 specifications [26].

The 3G cell is considered to be a circle with a radius of1000 meters, with a base station located in the center. It isassumed that 3G BCMCS supports a physical layer rate of204.8 kb/s, which is able to cover the entire cell using a(32, 28) Reed-Solomon error correction code, according to [1].In our simulation, the transmission power of the 3G basestation is set accordingly so that BCMCS can broadcast thevideo to the entire cell. The same base station transmissionpower is used in SV-BCMCS evaluations. The users’ averagePSNR is used as the metric of their received video quality.

TABLE IIRATE (KB/S) AND PSNR (DB) VALUES OF ALL THE LAYERS FOR THE

THREE SVC ENCODED VIDEO SEQUENCES

Mobile Football BusRate PSNR Rate PSNR Rate PSNR

Base layer 149.1 31.1756 136.0 30.0436 138.5 31.2975Enh. layer 1 237.5 31.9141 233.5 30.8987 217.3 31.9257Enh. layer 2 320.2 32.8918 353.5 32.8803 319.8 33.3549Enh. layer 3 391.8 33.9392 421.1 34.0621 422.0 35.3183Enh. layer 4 461.1 35.4830 506.4 36.1888 494.3 37.4306Enh. layer 5 522.0 37.1488 541.6 37.0958 545.1 39.1810

2) Scalable Video Settings: Three standard SVC test videosequences, Mobile, Football and Bus in QCIF resolution(176 × 144 pixels) with a frame rate of 15 frames/sec areused in the simulations. All of the sequences are availablefrom [27]. They are played repeatedly to yield video sequenceswith a length of approximately 90 seconds. Unless indicatedotherwise, the video length is the simulation length for mostof our simulation runs. We use JSVM 9.19.7 reference soft-ware to encode the video sequences into one base layer andfive SVC fidelity enhancement layers with MGS (Medium-Grain fidelity Scalability), based on the SVC extensions ofH.264/AVC [4]. In our setting, one GOP (Group-Of-Picture)includes 16 frames. By adjusting the quantization parameters(QP) for each layer in the encoding, all videos are encodedat the rate of about 530 kb/s. The resulting rate points andPSNR values for the layers of each encoded video sequencesare summarized in the Table II.

The generated packet information of each video sequenceis integrated into OPNET Modeler to simulate video transmis-sion and reception. Since we have modeled interference in OP-NET, there will be dynamic packet loss in the ad-hoc network.Thus the video layers are encoded independently using (6, 5)FEC to combat the packet loss, as described in Section III-C2.On the receiving side, a user device decodes an enhancementlayer if it has successfully received enough parity packets ofthis layer, otherwise this layer and all the higher enhancementlayers are discarded due to the decoding dependency. Finallywe use JSVM to decode the received stream for each user andmeasure the PSNR of the reconstructed video.

Note that although we can use any general non-decreasingutility function U(·), for simplicity, in our simulation wecharacterize the video stream by a set of PSNR points ul,which represent the PSNR of the encoded video with rencl =∑li=1Ri being the corresponding encoding rate of the scalable

video stream at layer l, l ∈ 1 . . . L. Then we solved the optimalresource allocation problem in the simulation with this model.The ul’s and rencl ’s are as listed in Table II.

B. Stationary Scenarios

In stationary scenarios, a certain number of fixed nodes(users) are uniformly distributed in the 3G cell. The presentedresults are averages over ten random topologies. The 90%confidence interval is also determined for each simulationpoint.

1) The Impact of Node Density: In this scenario, we usethe settings described in the previous section. The number of

9

0 100 200 300 400 500 600 700

32

32.5

33

33.5

34

Number of Multicast Receivers

Ave

rage

PS

NR

(dB

)

SV−BCMCS with ad−hoc SV−BCMCS without ad−hocTraditional BCMCS

(a)

0 100 200 300 400 500 600 70031.6

31.8

32

32.2

32.4

32.6

32.8

33

33.2

33.4

33.6

Number of Multicast Receivers

Ave

rage

PS

NR

(dB

)

SV−BCMCS with ad−hocSV−BCMCS without ad−hocTraditional BCMCS

(b)

0 100 200 300 400 500 600 70033.2

33.4

33.6

33.8

34

34.2

34.4

34.6

34.8

Number of Multicast Receivers

Ave

rage

PS

NR

(dB

)

SV−BCMCS with ad−hocSV−BCMCS without ad−hocTraditional BCMCS

(c)

Fig. 5. Impact of number of users (a) Video sequence Mobile (b) Video sequence Football (c) Video sequence Bus.

−0.5 0 0.5 1 1.5 2 2.5 3 3.5 4 4.531

31.5

32

32.5

33

33.5

34

34.5

35

Time−To−Live (TTL) of the ad−hoc network

Ave

rage

PS

NR

(dB

)

SV−BCMCS: MobileSV−BCMCS: FootballSV−BCMCS: BusBCMCS: MobileBCMCS: FootballBCMCS: Bus

Fig. 6. Impact of TTL on the average PSNR.

1 2 5 10 15 3030

30.5

31

31.5

32

32.5

33

33.5

34

Speed of the Mobile Nodes (Meter/sec)

Ave

rage

PS

NR

(dB

)

SV−BCMCS in Mobile ScenariosSV−BCMCS in Static Scenarios

Fig. 7. Impact of maxspeed.

120 130 140 150 160 17032.2

32.4

32.6

32.8

33

33.2

33.4

33.6

33.8

34

34.2

The Base Layer Encoding Rate (kb/s)

Ave

rage

PS

NR

(dB

)

SV−BCMCS with ad−hoc TTL = 3SV−BCMCS without ad−hoc

Fig. 8. Impact of base layer encoding rate.

multicast receivers in the cell is ranging from 100 to 600, tosimulate a sparse to dense node distribution. As a comparison,“Traditional BCMCS” or “BCMCS” indicate the transmissionof a single layer video, which we encode using JSVM single-layer coding mode. We encode the single-layer video intoapproximately 185 kb/s (Specifically, 180.6 kb/s for Mobile,184.9 kb/s for Football and 191.3 kb/s for Bus). The rateis close to the maximum rate allowed by the BCMCS withthe physical layer rate of 204.8 kb/s. The average PSNR forscenarios without an ad-hoc network, and scenarios with ad-hoc network (TTL set to 3) for all three video sequences, aregiven in Figure 5.

With and without ad-hoc data relay, SV-BCMCS consis-tently out-performs the traditional BCMCS. Without ad-hocrelay, SV-BCMCS provides approximately 0.8 dB (Mobilesequence), 0.6 dB (Football sequence) and 0.2 dB (Bus se-quence) gain, respectively. The ad-hoc relay leads to extraperformance gain. For example, for video sequence Bus, whenthe number of multicast receivers is 100, the ad-hoc relayinggives a 0.36 dB additional PSNR improvement. When thereceiver number is 600, the additional improvement reaches1.17 dB. In general, the PSNR gain with ad-hoc relay increasesas the number of users grows because more users facilitatead-hoc relaying compared to the traditional BCMCS with 600users, SV-BCMCS improves the users’ average PSNR by 1.70dB for Mobile sequence, and 1.70 and 1.35 dB for Footballand Bus sequences respectively.

2) The impact of number of relay hops (TTL): Figure 6depicts the performance of SV-BCMCS under different TTLs.The analysis in Section IV studies the users’ effective distancegain with the aid of ad-hoc data relay. Here the impactof ad-hoc relay is examined in a more practical setting: inthe presence of wireless interference and using real videosequences. The experiments are done with the number of usersset to be 500.

As shown in the figure, the performance of SV-BCMCSimproves as TTL increases. With more relaying hops, userscan potentially connect to the helpers closer to the 3G BS,thus obtaining more higher enhancement layers via ad-hocrelaying. For example, in the figure of Bus sequence, withTTL = 1, the average PSNR is about 0.66 dB higher than thatin the traditional BCMCS. Such an improvement increasesto 1.54 dB when the TTL reaches four. A similar trendcan be observed for the Mobile and Football sequences aswell. However, as the TTL becomes larger, the additionalinterference and communication overhead grow. Hence thePSNR curves flatten out gradually. Finally, note that even withTTL = 0, i.e., without ad-hoc relay, the SV-BCMCS still out-performs BCMCS due to the employment of SVC coding andthe optimal resource allocation.

C. Mobile Scenarios

The impact of user mobility to the users’ received videoquality, represented by PSNR, is studied next. The randomwalk model with reflection [28] is used to drive the user

10

TABLE IIIIMPACT OF RECONFIGURATION INTERVAL ON THE AVERAGE PSNR

(maxspeed=10 M/S)

ReconfigurationInterval (sec)

45 30 10 5

Users’ AveragePSNR (dB)

32.498 32.569 32.697 32.783

movement in the simulation. The individual user’s movingspeed is randomly selected in the range from zero to maxspeed(m/s), where maxspeed is a simulation parameter. Both movingspeed and moving direction are adjusted periodically, withthe time period drawn from a uniform distribution betweenzero and 100 seconds. We only present the results for theMobile sequence however similar observations are made forother video sequences. Also we loop the video into a longersequence with a length of 10 minutes. The mobility affectsSV-BCMCS’s performance in the following two ways: (i) theefficiency of ad-hoc network degrades due to the link failurescaused by mobility, and (ii) the optimal channel allocation isdisrupted since the user positions keep changing. SV-BCMCSperiodically reconfigures the optimal allocation so as to adaptto the user position change.

1) The Impact of moving speed: There are 600 users inthe cell, with ad-hoc relay hops (TTL) set to be three. Themaxspeed is set at 1, 2, 5, 10, 15 and 30 m/s, respectively. Thebase station reconfiguration interval of the optimal channelallocation is set to be 30 seconds. Figure 7 depicts theusers’ average PSNR for different maxspeed. Clearly, theperformance of the SV-BCMCS degrades as the users movefaster, especially when the speed increases beyond 5 m/s. Infact, when the speed is above 5 m/s, the users’ average PSNRis approaching the value without the aid of ad-hoc data relays,as shown in Section V-B.

2) The Impact of the Reconfiguration Interval: Table IIIsummarizes the impact of optimal allocation reconfigurationinterval on the average PSNR when maxspeed is 10 m/s. Asthe interval becomes shorter, the SV-BCMCS can adapt tothe mobile environment faster, leading to better performance.This is at the price of computational power, communicationoverhead, and changing video quality perceived by some users.Hence the reconfiguration interval should be selected to strikethe right balance.

D. Tradeoffs between the Base Layer Rate and the OverallPerformance

In SV-BCMCS, depending on a user’s location, it mayreceive the same video at different quality levels by receiving adifferent number of video layers. In the worst case, a user mayonly receive the base layer, which is transmitted to the entirecell. SV-BCMCS allows users having better channel conditionto receive better quality video, which is fair in the sense ofmaximizing the aggregate utility for all users. The study ofthe right fairness metric, however, is outside the scope ofthis paper. Here we focus on the tradeoffs between the baselayer rate and the overall improvement of user perceived videoquality.

Figure 8 depicts the average PSNR vs. the base layer rate inSV-BCMCS with and without ad-hoc data relay. The Mobilesequence is being used. There are 600 fixed users in the 3Gcell. We change the base layer rate by adjusting the QP forthe base layer during encoding. The QP for the enhancementlayer rate is fixed which gives a relatively similar encoding ratefor the enhancement layers. The resulting rates for base layerare 123.0, 136.1, 149.1 and 166.2 kb/s respectively. We cansee that as the base layer rate increases further beyond 136.1kb/s, the users’ average PSNR decreases. The base layer ratemust not exceed the single layer rate (204.8 kb/s) as used inBCMCS. Approaching this rate, the entire channel can onlytransmit the base layer, and there is hardly any differencebetween SV-BCMCS and BCMCS. Intuitively, the high baserate leaves less “room” for SV-BCMCS to optimize and toachieve a higher average PSNR.

Figure 9 shows the users’ average PSNR vs. the distance tothe base station in SV-BCMCS. The base layer rate is set to be149.1 kb/s and 166.2 kb/s respectively and all the other settingsare the same as above. Each point in the figure represents oneuser. Clearly, more users in the small base layer rate case(149.1 kb/s) are able to enjoy higher PSNR than in the largebase layer rate case (166.2 kb/s). Specifically, 301 users in thefirst case are with PSNR more than 32.0 dB, compared with208 users in the second case. Using less channel bandwidth todeliver a smaller rate base layer enables enhancement layersto be transmitted further. However, the users who only obtainthe base layer perceive worse video quality in the small baselayer rate case than in the large base layer rate case. With theaid of ad-hoc data relay, more users are able to receive videowith higher quality regardless of the base layer rate.

VI. DISCUSSION AND FUTURE WORK

In this section we discuss several issues relevant to the SV-BCMCS architecture.

Multiple Cells: The technique of soft-combining acrossmultiple cells for EV-DO BCMCS is proved to benefit the edgeusers by enhancing their data throughput and transmissionreliability. Currently we didn’t integrate the soft-combininginto our optimal resource allocation but only consider thesingle-cell case. This is left as our future work. With soft-combining enabled, the edge users will experience higher datarate, while the users in the middle of the cell can also havetheir multicast rate improved due to SV-BCMCS, which leadsto a more fair performance.

Opportunistic Receiving: In a real scenario, due to theeffect of fading and shadowing, the receiving signal strengthvaries in a small time scale, which results in variations inchannel conditions for each user. For example in Figure 2, itis possible that node A does not receive all the layer 4 packetswhile node B receives some of the packets from layer 4. Ournext step is to model the wireless links as probabilistic whichis approaching the reality. In this way, the optimal resourceallocation and relay routing will be redesigned accordingly.

Interference in Ad-hoc Network: Interference in the mainfactor limiting the throughput of multi-hop wireless ad-hocnetwork. Even without collision, simultaneous transmissions in

11

0 200 400 600 800 100031

32

33

34

35

36

37

38

Distance to the Base Station (Meters)

Ave

rage

PS

NR

(dB

)

SV−BCMCS nodesnot using ad−hoc

SV−BCMCS nodesusing ad−hocTraditional BCMCS

Base Layer Rate:149.1 kb/s

0 200 400 600 800 100031

32

33

34

35

36

37

38

Distance to the Base Station (Meters)

Ave

rage

PS

NR

(dB

)

SV−BCMCS nodesnot using ad−hocSV−BCMCS nodesusing ad−hocTraditional BCMCS

Base Layer Rate:166.2 kb/s

Fig. 9. Tradeoffs of the base layer rate and the overall performance in SV-BCMCS. Optimal resource allocation significantly improves the system-widevideo rate at the price of a small quality decrease for nodes close to the boundary; ad-hoc relays further increase the video quality for all users, almost allusers achieve a higher video quality than in the traditional BCMCS.

the same channel will interfere with each other, resulting in thedegradation of the link rates. The OPNET-based simulationswe conducted already include the interference effect. Whilesimulations give a fairly good performance, it still trails thetheoretical bound assuming perfect ad-hoc transmission, dueto interference. One of our future work is to optimize themulticast tree in ad-hoc relay network. Such multicast treeguarantees each user still receiving the video layers it shouldreceive decided in resource allocation, while minimizing thenetwork interference.

VII. CONCLUSION

In this paper we present SV-BCMCS, a novel scalable videobroadcast/multicast solution that efficiently integrates scalablevideo coding, 3G broadcast and ad-hoc forwarding. We for-mulate the resource allocation problem for scalable videomulticast in a hybrid network whose optimal solution canbe resolved by a dynamic programming algorithm. Efficienthelper discovery and video forwarding schemes are designedfor practical layered video/content dissemination through ad-hoc networks. Furthermore, we analyze the effective distancegain enabled by ad-hoc relay, which provides insight into thevideo quality improvement made possible by using ad-hoc datarelay. Finally, OPNET based real video simulations show thata practical SV-BCMCS increases the users’ average PSNR by1.35 ∼ 1.70 dB for the video sequences we use, with thead-hoc networks accounting for around 1.2 dB improvement.Moreover, SV-BCMCS still maintains a minimum of 0.80 dBperformance improvement when the nodes’ moving speed isless than 5 m/s, while periodical reconfiguration is necessaryin fast moving scenarios. The tradeoffs between the baselayer rate and the overall performance is discussed and wedemonstrate that SV-BCMCS can significantly improve thesystem-wide video quality, though a few viewers close to theboundary will have a slight quality degradation.

REFERENCES

[1] P. Agashe, R. Rezaiifar, and P. Bender, “CDMA2000 High Rate Broad-cast Packet Data Air Interface Design,” IEEE Commun. Mag., vol. 42,no. 2, pp. 83–89, February 2004.

[2] J. Wang, R. Sinnarajah, T. Chen, Y. Wei, and E. Tiedemann, “Broadcastand Multicast Services in CDMA2000,” IEEE Commun. Mag., vol. 42,no. 2, pp. 76–82, February 2004.

[3] “Third Generation Partnetship Project (3GPP2).” [Online]. Available:http://www.3gpp2.org/

[4] H. Schwarz, D. Marpe, and T. Wiegand, “Overview of the ScalableVideo Coding Extension of the H.264/AVC Standard,” IEEE Trans.Circuits Syst. Video Technol., vol. 17, no. 9, pp. 1103–1120, September2007.

[5] H. Luo, R. Ramjee, P. Sinha, L. E. Li, and S. Lu, “UCAN: AUnified Cellular and Ad-Hoc Network Architecture,” in Proc. of ACMMOBICOM, 2003.

[6] H. Hsieg and R. Sivakumar, “On Using Peer-to-Peer Communication inCellular Wireless Data Networks,” IEEE Trans. Mobile Comput., vol. 3,no. 1, pp. 57–72, February 2004.

[7] J. C. Park and S. K. Kasera, “Enhancing Cellular Multicast PerformanceUsing Ad Hoc Networks,” in Proc. of IEEE Wireless Comm. andNetworking Conf. (WCNC), 2005.

[8] R. Bhatia, L. E. Li, H. Luo, and R. Ramjee, “ICAM: Integrated Cellularand Ad Hoc Multicast,” IEEE Trans. Mobile Comput., vol. 5, no. 8, pp.1004–1015, August 2006.

[9] K. Sinkar, A. Jagirdar, T. Korakis, H. Liu, S. Mathur, and S. Panwar,“Cooperative Recovery in Heterogeneous Mobile Networks,” in Proc.of IEEE SECON, 2008.

[10] C.-S. Hwang and Y. Kim, “An Adaptive Modulation Method forMulticast Communications of Hierarchical Data in Wireless Networks,”in Proc. of IEEE International Conf. on Commun. (ICC), 2002.

[11] J. Shi, D. Qu, and G. Zhu, “Utility Maximization of Layered Video Mul-ticasting for Wireless Systems with Adaptive Modulation and Coding,”in Proc. of IEEE International Conf. on Commun. (ICC), 2006.

[12] J. Kim, J. Cho, and H. Shin, “Layered Resource Allocation for VideoBroadcasts over Wireless Networks,” IEEE Trans. on Consumer Elec-tronics, vol. 54, no. 4, pp. 1609–1616, November 2008.

[13] P. Li, H. Zhang, B. Zhao, and S. Rangarajan, “Scalable video multicast inmulti-carrier wireless data systems,” in Network Protocols, 2009. ICNP2009. 17th IEEE International Conference on, 2009.

[14] D. Hu, S. Mao, Y. Hou, and J. Reed, “Scalable video multicast incognitive radio networks,” Selected Areas in Communications, IEEEJournal on, vol. 28, no. 3, pp. 334 –344, april 2010.

[15] T. Schierl, S. Johansen, A. Perkis, and T. Wiegand, “Rateless scalablevideo coding for overlay multisource streaming in manets,” J. Vis.Commun. Image Represent., vol. 19, no. 8, pp. 500–507, 2008.

[16] S. Hua, Y. Guo, Y. Liu, H. Liu, and S. Panwar, “Sv-bcmcs: Scalablevideo multicast in hybrid 3g/ad-hoc networks,” in Global Telecommuni-cations Conference, 2009. GLOBECOM 2009. IEEE, nov. 2009.

[17] K. Stuhlmuller, N. Farber, M. Link, and B. Girod, “Analysis of videotransmission over lossy channels,” Selected Areas in Communications,IEEE Journal on, vol. 18, no. 6, pp. 1012 –1032, 2000.

[18] H. Zhang, Y. Zheng, M. Khojastepour, and S. Rangarajan, “Scalablevideo streaming over fading wireless channels,” in Wireless Communi-cations and Networking Conference, 2009. WCNC 2009. IEEE, 2009.

[19] M. van der Schaar, S. Krishnamachari, S. Choi, and X. Xu, “Adaptivecross-layer protection strategies for robust scalable video transmissionover 802.11 wlans,” Selected Areas in Communications, IEEE Journalon, vol. 21, no. 10, pp. 1752 – 1763, dec. 2003.

[20] S. Hua, Y. Guo, Y. Liu, H. Liu, and S. S. Panwar, “Scalable VideoMulticast in Hybrid 3G/Ad-hoc networks,” Polytechnic Institute of New

12

York University, Tech. Rep., March 2009, http://eeweb.poly.edu/faculty/yongliu/docs/SV-BCMCS tech.pdf.

[21] W.-T. Tan and A. Zakhor, “Video multicast using layered fec andscalable compression,” Circuits and Systems for Video Technology, IEEETransactions on, vol. 11, no. 3, pp. 373 –386, mar 2001.

[22] S. Mao, S. Lin, S. Panwar, Y. Wang, and E. Celebi, “Video transportover ad hoc networks: multistream coding with multipath transport,”Selected Areas in Communications, IEEE Journal on, vol. 21, no. 10,pp. 1721 – 1737, dec. 2003.

[23] M. Fewell, “Area of Common Overlap of Three Circles,” Tech. Rep.DSTO-TN-0722, 2006. [Online]. Available: http://hdl.handle.net/1947/4551

[24] “OPNET.” [Online]. Available: www.opnet.com[25] P. Bender, P. Black, M. Grob, R. Padovani, N. Sindhushayana, and

A. Viterbi, “CDMA/HDR: A Bandwidth Efficient High Speed WirelessData Servicefor Nomadic Users,” IEEE Commun. Mag., vol. 38, no. 7,pp. 70–77, July 2000.

[26] “CDMA2000 High Rate Broadcast-Multicast Packet Data Air InterfaceSpecification,” 3GPP2 Specifications, C.S0054-A v1.0 060220, Febru-ary 2006.

[27] “SVC test sequences.” [Online]. Available: ftp.tnt.uni-hannover.de/pub/svc/testsequences

[28] T. Camp, J. Boleng, and V. Davies, “A Survey of Mobility Modelsfor Ad Hoc Network Research,” Wireless Communications and MobileComputing (WCMC), vol. 2, no. 5, pp. 483–502, 2002.

Sha Hua received the B.S. degree from the Depart-ment of Electronics and Information Engineering,Huazhong University of Science and Technology(HUST), Wuhan, China in 2007. He is currentlypursuing the Ph.D. degree at the Department ofElectrical and Computer Engineering of PolytechnicInstitute of New York University, Brooklyn, NY.

His research interests include the optimizationand protocol design in wireless video transmission,next-generation cellular system and cognitive radionetworks.

Yang Guo is a researcher with Service InfrastructureResearch Department at Bell Labs, Alcatel-Lucent,at Crawford Hill, Holmdel, New Jersey. Prior to thecurrent position, he was a Principal Scientist withCorporate Research, Technicolor (formerly Thom-son). Yang has a strong background in the fieldof distributed systems, multimedia, and networking.His research interests lie in multimedia systemsand networking in general, and peer-to-peer, contentdistribution, online social networks, IPTV, and mediastreaming in particular. Yang Guo obtained his Ph.D.

from University of Massachusetts at Amherst, and M.S. and B.S. fromShanghai JiaoTong University.

Yong Liu has been an assistant professor at theElectrical and Computer Engineering department ofthe Polytechnic Institute of NYU since March, 2005.He received his Ph.D. degree from Electrical andComputer Engineering department at the Universityof Massachusetts, Amherst, in May 2002. He re-ceived his master and bachelor degrees in the fieldof automatic control from the University of Scienceand Technology of China, in July 1997 and 1994respectively. His general research interests lie inmodeling, design and analysis of communication

networks. His current research directions include robust network routing, Peer-to-Peer IPTV systems, overlay networks and network measurement. He is thewinner of the IEEE INFOCOM Best Paper Award in 2009, and the IEEECommunications Society Best Paper Award in Multimedia Communicationsin 2008. He is a member of IEEE and ACM.

Hang Liu received the Ph.D. degrees in electricalengineering from the University of Pennsylvania in1997. From 1997 to 2000, he was with NEC Labo-ratories America, Princeton, NJ, working on one ofthe first pre-commercial 5 GHz broadband wirelesssystems. From 2000 to 2003, he worked for twostartup companies, Iospan Wireless, Inc. and TelliumOptical Systems, Inc., designing and developing net-work management and network security systems forbroadband wireless access and IP-centric networkcontrol and routing platform for switched optical

networks, respectively. From 2004 to 2010, he is with Corporate ResearchLab, Thomson Inc, Princeton, NJ, where he is a principal scientist and projectmanager, led a team to conduct research in the areas of wireless networking,wireless video streaming, and content distribution. He has also been an adjunctprofessor of WINLAB, the ECE department, Rutgers University since 2004.He has published more than 50 papers in the areas of wireless communicationsand networking, wireless/Internet multimedia communications systems, andIP-based optical networks, and was a co-recipient of one best paper awardand one best student paper award. He is the inventor/co-inventor of over 50granted and pending international patents. He has also actively contributed tothe IEEE 802 wireless standards including IEEE 802.11aa, 802.11s, 802.16jand 802.22.

Shivendra S. Panwar is a Professor in the Electricaland Computer Engineering Department at Polytech-nic Institute of NYU. He received the B.Tech. degreein electrical engineering from the Indian Instituteof Technology, Kanpur, in 1981, and the M.S. andPh.D. degrees in electrical and computer engineeringfrom the University of Massachusetts, Amherst, in1983 and 1986, respectively.

He joined the Department of Electrical Engi-neering at the Polytechnic Institute of New York,Brooklyn (now Polytechnic Institute of New York

University). He is currently the Director of both the New York State Center forAdvanced Technology in Telecommunications (CATT) and the NSF sponsoredWireless Internet Center for Advanced Technology (WICAT). He spent thesummer of 1987 as a Visiting Scientist at the IBM T.J. Watson ResearchCenter, Yorktown Heights, NY, and has been a Consultant to AT&T BellLaboratories, Holmdel, NJ. His research interests include the performanceanalysis and design of networks. Current work includes cooperative wirelessnetworks, switch performance and multimedia transport over networks.

He is an IEEE fellow and has served as the Secretary of the TechnicalAffairs Council of the IEEE Communications Society. He is a co-editorof two books, Network Management and Control, Vol. II, and MultimediaCommunications and Video Coding, both published by Plenum, and hasco-authored “TCP/IP Essentials: A Lab based Approach” published by theCambridge University Press. He was awarded, along with Shiwen Mao,Shunan Lin and Yao Wang, the IEEE Communication Society’s Leonard G.Abraham Prize in the Field of Communication Systems for 2004.